Multimodal Mental Health Modeling Toolbox

​

This toolbox was developed to support research in computational psychiatry, particularly for integrating and analyzing multimodal data in predictive modeling of mental health outcomes. It was used in the experiments for our study on latent space fusion methods for depression symptom prediction using the Brighten dataset.

 

Features

• Preprocessing:

  • Missing data handling with methods like MissForest and statistical imputation.

  • Feature scaling and normalization.

  • Categorical encoding (One-Hot, Label Encoding).

  • Modular merging and filtering of multimodal features.

• Data Synchronization:

  • Temporal alignment of time-series data from different modalities (e.g., GPS, survey, phone logs).

  • Resampling and interpolation to handle asynchronous data streams.

• Modeling:

  • Support for regression and classification tasks.

  • Trained models include:

    • Random Forest

    • Linear & Logistic Regression

    • SVM

    • Neural Networks (MLP)

  • Simple API for training, evaluation, and comparison.

• Utilities:

  • Model persistence (save/load via joblib).

  • Performance logging.

  • Generic file loading and encoding helpers.

 

Example Use

 

 

 

 

 

 

 

 

 

 

 

 

 

Modules

​

​

​

​

​

​

​

​

​

​

 

Research Context

This toolbox was used in the analysis of smartphone-sensed behavioral, clinical, and demographic data from the Brighten clinical trial. It enabled the comparison of early fusion and latent space fusion techniques for predicting depressive symptom severity (PHQ-2 scores).

​

Disclaimer

This code is provided as-is for research and educational purposes only. It uses open-source libraries (e.g., scikit-learn, pandas, numpy) and was developed in an academic context. It is not intended for clinical use or deployment in production systems. No warranty is provided for its performance or correctness.

​

Interested in our toolbox?
Explore it here: https://github.com/McPsyt/MMHM-ToolBox